[ovs-discuss] VM's on different host are not able to communicate.

[Jesse Gross]

On Tue, Jan 15, 2013 at 4:01 AM, Anil Vishnoi <vishnoianil at gmail.com> wrote:
> Hi,
>
> I have a setup of 2 server connected through one L2 Rack switch. Ubuntu
> 12.04 + KVM + openvswitch is installed on both the servers. I created one VM
> each on both the server. IP address of both the host and 2 VM's are in the
> same subnet (10.10.*.*/16). eth1 of hosts is connected to their respective
> openvswitch bridge. VM's vnet interface is also connected to the openvswitch
> bridge. openvswitch is set for standalone fail-safe mode.
> when i ping server 1 to server 2, ping works fine and i can see the flows in
> their respective bridge using ovs-dpctl. I can ping from the host to the VM
> hosted on the same server as well. But when i log-in to the VM and try to
> ping the other host or VM on other host it doesn't work. Reason is because
> of failure of ARP resolution. When i ping host1-Vm to host2, i can see that
> arp request is reaching host 2 and host 2 replied with the ARP response, but
> that response is not reaching back to host1-VM. If i setup static ARP entry
> then i can see that ICMP request reach to the host 2 and also it sends back
> reply, but that reply is not reaching to the host1-VM, so its ideally the
> same scenario as i saw in ARP resolution. Ideally i should see response
> packet (ARP/ICMP) on eth1 of host1 in tcpdump, but i am not seeing those
> packet in tcpdump output. Is it the case that openvswitch is dropping those
> packets and hence tcpdump is not showing it. I even ran tcpdump on eth1 with
> promiscuous mode, but nothing changed.

tcpdump runs before Open vSwitch so even if the packet is dropped it
should still show up.  Do you see the packet on the outgoing physical
interface?

In this case, there is very little difference between VMs and the host
from Open vSwitch's perspective.